Traffic Prophet

Traffic Prophet is a suite of Python tools to transform Toronto's historical traffic count data to estimates of annual average daily traffic (AADT) on all city streets. It is based on the Traffic Emissions Prediction Scheme (TEPS) codebase created by Dr. Arman Ganji of the University of Toronto (paper).

Traffic Prophet currently remains under development. On initial release, it will contain:

• A method, based on Bagheri et al. 2014, of extrapolating short-term traffic counts into AADTs using nearby long-term counts.
• Gaussian process regression method for estimating arterial AADTs from nearby observed AADTs.
• Support vector regression method for estimating local road AADTs from nearby observed AADTs.

Alongside the Traffic Prophet codebase, this repo also contains some literature review, a number of development and experimental notebooks in a sandbox folder, utilities for working with TEPS and scripts for helping to generate input data for the model.

Folder Structure

• input_data - scripts to produce input data tables on Postgres.
• lit_review - literature review, notably for Bagheri et al. 2013, the basis for CountMatch.
• sandbox - Developmental and experimental notebooks in support of building out TEPS.
• teps - tutorials on how to run TEPS, and how to bootstrap components of Traffic Prophet to produce input datasets for TEPS.
• traffic_prophet - Traffic Prophet codebase. See below for details.

Traffic Prophet Codebase

• connection.py - basic wrapper for storing Postgres connections.
• config_template.py - template for creating config.py.
• countmatch/ - port of the TEPS version of the Bagheri et al. 2014 algorithm.
  • base.py - base classes and shared code.
  • derivedvals.py - aggregate values and weights, such as MADT, and DoM_ijd (ratio between MADT and day-to-month ADT).
  • growthfactor.py - fit for year-on-year multiplicative growth factor.
  • matcher.py - matcher between short and permanent counts, and AADT estimator classes.
  • neighbour.py - nearest neighbour calculator.
  • permcount.py - find which counts in a dataset have enough data to be considered permanent counts.
  • reader.py - reads
  • tests/ - test suite for countmatch
    • conftest.py - common components of tests.
    • test_{countmatch_file}.py - test suite for {countmatch_file}.
• data/ - sample data used in test suites. See __init__.py for definitions.
• tests/ - integration testing for the entire volume model. Currently empty.

Requirements

As given by requirements.txt, the Traffic Prophet codebase requires:

hypothesis>=5.5.4
numpy>=1.18
pandas>=1.1
psycopg2>=2.8.4
pytest>=5.4
scikit-learn>=0.22
statsmodels>=0.11.1
tqdm>=4.43

Individual Jupyter notebooks in sandbox may require other packages (including notebook to actually run Jupyter notebooks, of course).

Usage

Installation

traffic_prophet itself only requires a Python environment with the packages listed above. It, however, also relies on input data from Postgres (or zip files, but this is a legacy ingestion method developed for testing against TEPS). The required Postgres tables are:

prj_volume.tp_centreline_lonlat.sql
prj_volume.tp_centreline_volumes.sql
prj_volume.tp_daily_volumes.sql

The scripts to create these are located in input_data/flow.

traffic_prophet also assumes there to be a user-generated config.py file inside the traffic_prophet/ directory. Also in that directory is a config_template.py file that shows how to create config.py.

For a default configuration, simply copy the contents of config_template.py into a new config.py file.

Importing

To import traffic_prophet, add this folder to the Python PATH, eg. with

import sys
sys.path.append('<FULL PATH OF bdit_traffic_prophet>')

Testing

To test the package, run the following in this folder:

pytest -s -v --pyargs traffic_prophet

License

Traffic Prophet is licensed under the GNU General Public License v3.0 - see the LICENSE file.